Ucal, Meltem Şengün2019-06-272019-06-2720050978-980-6560-60-4https://hdl.handle.net/20.500.12469/141The most important stage of econometrics estimation is the in model set up. The model set up has the best prediction ability and is therefore suitable for econometrics estimation. Between the dependent variable Y and other independent explanatory variables X must be a strong relationship in econometrics estimation. However all explanatory variables cannot be related to dependent variable Y. This condition creates a regression problem. A similar problem appears in variable selection equivalent to problem in model selection. The most suitable faultless model is provided by correct and suitable selection of variables. There exist many variable/model selection procedures the where the necessary relationship between X and Y is linearfor example the C-P methodthe Bayes Information Criterion (BIC) (Hannan-Quin)the Final Prediction Error Method (FPE lambda - Shibata. 1984)Akaike Information Criterion (ACI)Schwartz Criterion (SC)Cross-Validation (CV)Generalized Cross Validation (GCV) (Craven-Wahba)Log LikelihoodBootstrap and Jackknife. In this paper we compare some different model selection criteria with the parametric bootstrap and present a simple procedure to obtain a linear approximation of the mean squared prediction error. This study is based on empirical evidence and model training.eninfo:eu-repo/semantics/closedAccessModel selection criteriaBootstrap selection criterionCandidate for true modelMean squared prediction errorBootstrap estimator of the expected excess errorConference Object354357WOS:0002436876000682-s2.0-84867374925N/AN/A